Meet Your AI Mahjong Coach

Three powerful tools to transform your mahjong skills, powered by deep reinforcement learning.

Real-Time Hand Analysis

Input your current hand and get instant AI recommendations:

  • Best discard with confidence percentage
  • Win rate estimation for your current hand
  • Alternative plays with risk/reward tradeoffs
  • Reasoning — understand why, not just what
  • 12ms inference — faster than you can blink
💡 Example
Hand: 1m 2m 3m 4p 5p 6p 7s 8s 9s 1z 1z 3z 5z
AI says: Discard 5z (73.2% confidence)
Reason: "Isolated honor tile with zero connectivity. Removing it keeps your three-suited structure intact."
Try Hand Analysis →
📊

Post-Game Review

Upload your game log and get a comprehensive analysis:

  • Move-by-move grading — Good / Inaccuracy / Mistake / Blunder
  • Critical moments highlighted with AI suggestions
  • Defensive analysis — were you dealt into a hand?
  • Efficiency tracking — how well did you progress toward tenpai?
  • Overall accuracy score — track improvement over time
📈 Like a Chess Engine
Think Stockfish, but for mahjong. Our AI evaluates every decision you made and shows you the optimal play.
Coming in Beta 🔜
🧠

Daily Quiz Training

5 hand scenarios daily to sharpen your decision-making:

  • Discard selection — choose the best tile to discard
  • Defense drills — identify dangerous tiles
  • Calling decisions — when to chi/pon/kan
  • Riichi timing — when to declare vs dama
  • Leaderboard — compete with other learners
🏆 Gamified Learning
Earn XP, maintain streaks, unlock achievements. Learning mahjong has never been this fun.
Coming in Beta 🔜
Under the Hood

Our AI is built on state-of-the-art deep reinforcement learning.

PPO
Algorithm
1M+
Self-Play Games
85%
DDZ Win Rate
63%
Changsha Win Rate

🔬 How Our AI Trains

1
Self-Play

Four AI agents play millions of games against each other, exploring strategies no human has tried.

2
PPO Optimization

Proximal Policy Optimization refines the AI's strategy, balancing exploration and exploitation.

3
Reward Shaping

Potential-based reward shaping guides learning with intermediate signals — not just win/lose.

4
Continuous Training

Models train 24/7 on cloud GPUs, getting stronger every day. Current training: 500K+ episodes per variant.

Ready to Try the AI Coach?

Free during beta. No download, no signup required to try.

Try It Now → Apply for Beta